Congressional Agenda Control and the Decline of Bipartisan Cooperation Laurel Harbridge Northwestern University College Fellow, Department of Political Science College Fellow, Institute for Policy Research l-harbridge@northwestern.edu Evidence shows that polarization in Congress has increased substantially since the 1970s while polarization in the public has increased much less, if at all. These two patterns of polarization suggest that responsiveness by members to their constituents has declined. By breaking apart congressional behavior, however, this paper suggests that this presumption is misleading. Looking at roll call votes and bill cosponsorship coalitions, this paper suggests that although partisan behavior has increased substantially in roll call votes, the same is not true for bill cosponsorship coalitions. In turn, this suggests that while responsiveness has declined when considering roll call voting, responsiveness has increased when considering cosponsorship coalitions. These divergent patterns can be reconciled by taking into consideration congressional agenda control.
[The] resulting polarization is worrisome. By reducing the space for bipartisanship, it can condemn Congress to gridlock. By driving elected officials to the fringe while citizens inhabit the center, it can alienate citizens from their government (Washington Post "A Polarized Nation?" 2004). From Congress to the airwaves to the bestseller lists, American politics appears to be hardening into uncompromising camps, increasingly identified with the two parties As it becomes more difficult to reach across the party line, campaigns are devoting more energy to firing up their hard-core supporters. For voters in the middle, this election may aggregate their feeling that politics no longer speaks to them, that it has become a dialogue of the deaf, a rant of uncompromising extremes (Van Drehle 2004). There is a presumption in the literature both by journalists and academics that party polarization has contributed to declining responsiveness by members of Congress to the public. Evidence shows that polarization in Congress has increased substantially since the 1970s while polarization in the public has increased much less, if at all. It seems that the logical next step is the assumption that responsiveness has declined. By breaking apart congressional behavior, however, this paper suggests that this presumption is misleading. Looking at roll call votes as well as bill cosponsorship coalitions, this paper suggests that although partisan behavior has increased substantially in roll call votes, the same is not true for bill cosponsorship coalitions. In turn, this suggests that while responsiveness has declined when considering roll call voting, responsiveness has increased when considering cosponsorship coalitions. These divergent patterns can be reconciled by taking into consideration congressional agenda control. Although bipartisan cooperation in voting has significantly declined, a pattern that will be documented in the following section, bipartisan cooperation in bill cosponsorship coalitions has not seen a similar sized decline. Thus, there appears to be an equilibrium level of bipartisan cooperation in Congress. However, the form that bipartisan cooperation has taken has changed over time as a result of changes in congressional agenda control. When bipartisan legislation became decreasingly likely to face roll call votes in the 1980s and 1990s, bipartisan cooperation 1
on bill cosponsorship coalitions increased. I argue that this pattern reflects pressures on members from more competitive districts to show they are moderate and willing to work across the aisle, and may suggest that members from opposing parties have more common ground than is often acknowledged. By bringing an explicit focus to the congressional roll call agenda, this paper examines how party control over the congressional agenda, and specifically the temporal changes in the selection of bills to receive roll call votes, affects the degree of bipartisan cooperation in Congress. Where we look for bipartisanship, and how we measure legislative behavior and preferences determines whether or not we see Congress as becoming increasingly polarized and partisan, and ultimately, whether we see members as out of step with their constituents. These results have a number of implications for understanding polarization in American politics, agenda control in congressional policy-making, and highlight the importance of carefully considering the potential biases in roll call data. An exclusive focus on roll call votes to measure bipartisanship or party polarization may misleading and can either under or overestimate partisanship, depending on how the congressional agenda is structured. Thus, these results suggest that claims of party polarization (at least in terms of the underlying ideologies of members) are over-stated. While it is clear that voting patterns have changed since the 1970s, consideration of the agenda and the selection of bills to face roll call votes must be taken into account before attributing changes to the underlying ideologies of members. Certainly the ideologies of members matter both in cosponsorship and voting but attention must also be given to how the parties structure the agenda. Literature on political parties has often debated when and how parties are influential, particularly in weak party systems such as the United States (Cox and McCubbins 2005; Krehbiel 1993, 2006). The findings in this paper are similar 2
to the findings by Cox and McCubbins, suggesting that party influence occurs through congressional agenda control. Rather than focusing on the ultimate degree of party unity on votes, however, this paper focuses on the changing likelihood over time that bipartisan bills receive roll call votes. From this perspective, much of the party polarization observed on roll call votes is an artifact of congressional agenda control and may not necessarily reflect the underlying preferences or ideologies of individual members. Similar to Fiorina s (2004) finding of similarities in public opinion and values between citizens in red and blue states, my findings suggest that this common ground may extend to members of Congress to a greater degree than is generally recognized. When we assess legislative behavior absent agenda control, members of Congress continue to engage in bipartisanship. Members of Congress have become increasingly representative of their districts when using bill cosponsorship coalitions, despite the fact that representation has declined when using roll call votes. Methodologically, this work highlights some of the problems in the standard results of roll call data to assess the preferences of individual members as well as trends in party polarization. By focusing exclusively on this stage of the legislative process, one that is strongly determined by the choice of which bills face roll call votes, these measures may overstate the differences of rank-and-file party members. This paper proceeds as follows. Section 2 sets out the groundwork for the argument that party polarization has increased, bipartisan cooperation has declined, and thus, that electoral responsiveness has declined. Section 3 breaks apart legislative behavior from the typical reliance on voting patterns, considering members bill cosponsorship coalitions. Section 4 reconciles the divergent patterns of bipartisan cooperation in voting and bill cosponsorship by considering the strategic use of agenda control. Section 5 concludes. 3
2. Party Polarization and the Decline of Electoral Responsiveness Half a century ago, political scientists yearned for distinctive parties (APSA 1950) but the resurgence of polarized parties since the 1970s has left few satisfied. Though there are some that continue to defend partisanship and polarization (e.g., Muirhead 2006; Rosenblum 2008), the more common refrain is that party polarization has come at the expense of an idealized era of bipartisanship (e.g. Eilperin 2006). Combined with evidence that the public has not polarized (Fiorina et al. 2004; DiMaggio et al. 1996), this body of research suggests that there is a disconnect between the mass public and the government (Fiorina and Levendusky 2006). Ultimately, this suggests that collective or aggregate electoral responsiveness has declined. This section summarizes the prevailing argument, drawing on the voting patterns of members. 1 Section 2.1 Increasing Polarization and Declining Bipartisanship The standard measures of polarization as well as legislative behavior come from the analysis of roll call votes in Congress. This section considers various measures of polarization and bipartisan cooperation, all of which rely on roll call votes themselves or the subsequent ideology estimates derived from votes. Let us begin with trends in party polarization. Regardless of the measure that is used, party polarization has significantly increased since the 1970s. Consider one common measure, the difference in party means, using either the first dimension of DW-NOMINATE or the Americans for Democratic Action (ADA) scores of roll call votes. While the former uses all non-unanimous votes and the latter uses only a small sample of key votes, the overtime patterns 1 The analysis focuses on the period from 1973 to 2004. This choice was driven by the visibility of voting decisions after the House reforms of the 1970s (Rohde 1991) and subsequent tracking of member s voting patterns by interest groups, and the use of cosponsorship in the House being restricted prior to the late-1960s. The House of Representatives was chosen because it is where party influence is often thought to be the greatest and where bipartisan cooperation is thought to be lowest (in comparison with the Senate, where bipartisanship is often thought to occur to a greater degree). 4
are the same. Polarization has increased nearly monotonically since the 1970s. Figure 1 presents the trend in polarization since 1973 using ADA scores for the House of Representatives. A second measure of partisanship considers the incidence of party unity votes in Congress. Congressional Quarterly (CQ) defines party unity votes as roll call votes that split the parties, a majority of voting Democrats opposing a majority of voting Republicans ("CQ Fact Sheet Bipartisan Voting" 1970, 1139). Since the discussion of cosponsorship coalitions will focus on the frequency of bipartisan behavior, it is useful to consider the flip side of party unity votes, or what CQ classifies as bipartisan votes. Bipartisan votes are roll-call votes on which a majority of voting Democrats and a majority of voting Republicans agreed ("CQ Fact Sheet Bipartisan Voting" 1963, 735). Since bills on which either party divides evenly are excluded, the percentage of bills that are bipartisan by CQ s measure is the complement of Party Unity roll calls (100 - % of Party Unity Votes). Since the CQ measure classifies all votes (except those that are omitted that exactly split a party) as either party unity or bipartisan votes, partisan and bipartisan are, by assumption, definitionally opposite. The CQ Bipartisan Measure (see Figure 2) for the House of Representatives shows a fairly steady decline in bipartisanship from the 1973 to 1995 with some recovery of bipartisanship beginning in the 105 th Congress (post-1996). 2 [Figure 2 about here] Since the Congressional Quarterly measure of bipartisanship looks at what majorities of 2 A similar pattern is found when looking just at final passage votes and omitting procedural issues. A number of scholars have noted that procedural votes have become increasingly partisan, either due to party pressure or good teamsmanship among members (Theriault 2006; Lee 2009). To check whether the relationship between polarization and bipartisanship holds when procedural votes are excluded, I draw on data compiled by David Rohde (2004) for House roll call voting. Looking only at final passage votes and only those on House bills (rather than resolutions or amendments), I find that the percentage of roll call votes that are bipartisan (using the CQ definition) declined from approximately 80 percent in 1973 to 40 percent by 1994. Although there was an increase in bipartisanship under the Republicans, the overall trend of the series has been a strong decline in bipartisanship consistent with the aggregate roll call results. Plots of the trends using final passage votes are available in the appendix. 5
the two parties are doing, it may miss additional and important variation. In particular, the CQ measure misses the extent of skew within voting (i.e., is it 90% of one party against 90% of another party or 52% of one party against 48% of another party). To parse out this element in the analysis of bipartisanship, I calculate a Difference in Party Support Score for each vote, focusing on the Yea votes. 3 For each vote, I calculate the percent of voting Democrats voting Yea and subtract the percent of voting Republicans voting Yea. The resulting score ranges from -100 to 100, where -100 indicates a vote on which no Democrats voted Yea and all of the Republicans voted Yea. The midpoint of 0 indicates bills where the same percentage of Democrats and Republicans voted Yea. Thus, the ends of scale reflect the greatest partisanship and the middle represents the greatest bipartisanship. Omitting unanimous and near unanimous (90% or more in favor) votes, I first plot the density of the scale by Congress for Yea votes when all House roll calls are included (Figure 3). 4 The trend over time has been an increase in density at the two poles of the distribution at the expense of the middle, or of the most bipartisan bills. Restricting the analysis to final passage votes on House bills also shows a decline in bipartisanship (see Appendix Figure 2). In contrast to the analysis of all votes, however, the final passage votes on the Yea side are skewed toward the majority party (the Democrats prior to the 104 th Congress and the Republicans from the 104 th forward). This skew is evidence of agenda control; the majority party is able to control the issues that came to votes in a way that allowed them to be the dominant group of proponents. In sum, bipartisan cooperation on roll call voting shows significant declines across the last three decades, 3 The analysis focuses on Yea votes since they are most similar to cosponsorship coalitions in indicating support for a piece of legislation. 4 When unanimous and near unanimous votes are included, the vast majority of legislation is bipartisan. This further suggests that roll call vote analyses, which tend to omit unanimous and near unanimous votes because they do not provide additional information for distinguishing between the preferences of various members, are biased against finding bipartisan cooperation. 6
particularly from the 1970s through the mid-1990s. This decline in bipartisanship is consistent with the rise in elite polarization. [Figure 2 about here] Section 2.2 Polarization in the American Public? Despite the abundance of evidence that Congress has become more polarized and partisan, there is significantly more skepticism that the public has grown as polarized. For instance, DiMaggio et al. (1996) found little evidence that the American publics social attitudes have become more polarized over the previous two decades, with the exception being attitudes on abortion. Similarly, Fiorina (2004) found little evidence of growing disparities between the opinions of red and blue state voters. While political elites including party activists have polarized, the general electorate has not shifted its positions. The electorate only appears polarized because it is choosing between the alternatives set by the (polarized) political elite (Fiorina et al. 2004). Section 2.3 Polarization and Declining Responsiveness Although political scientists often argue that concerns about reelection drive members to be representative of and responsive to their constituents, the dual patterns of increased elite polarization and little mass public polarization suggest that members are achieving reelection without being representative of their constituents. This breakdown in representation can be seen both in the relationship between legislators voting scores and summary measures of district preferences, as well as from survey responses regarding cooperation and bipartisanship. Scholars have long argued that reelection is a driving, if not the driving, force behind the behavior of members of Congress (Mayhew 1974; Fenno 1978). Normatively, the electoral connection suggests that both sides win; members of Congress get what they want (to stay in 7
office), and the public gets what they want (representative and responsive members). Taken to the extreme, Anthony Downs (1957) argued that in a two-party system, the candidates should converge completely to the median voter in the district. The Downsian framework also predicts centrism and responsiveness in members responses to the preferences of voters (Achen 1978). A large body of work has grown around the question of representation and whether members converge to the position of the district median what would be perfect representation (e.g., Ansolabehere et al. 2001; Burden 2004). Despite findings that show that candidates rarely fully converge in their positions, scholars have maintained that the median voter is important and that members can be electorally punished if they stray too far from the districts preferences (Canes-Wrone et al. 2002). Recent work has focused more specifically on the electoral impact of party cohesion, finding that members are punished electorally for being too partisan for the district and having high party unity support scores (Carson et al. 2010). Additional evidence that the public wants members who are less polarized, and are more bipartisan, comes from a recent Pew survey. A May, 2010 poll found that a substantial minority (42%) say they would be more likely to vote from a candidate who will make compromises with people they disagree with; only about half as many (22%) say they would be less likely to back a candidate willing to compromise (Pew 2010). Logically, it appears that a polarized Congress without equivalent polarization in the public damages representation. What does the data suggest? To begin, let s select a measure of district preferences, with which to judge legislative behavior in relation to. This paper draws on the Normal Presidential Vote in the district, measured as the mean two party presidential vote in 8
the previous two elections by the party of the Representative (Canes-Wrone et al. 2002). 5 For instance, if the member is a Republican I use the mean Republican presidential vote in the last two presidential elections and if the member is a Democrat I use the mean Democratic presidential vote in the last two presidential elections. 6 The presidential vote has been found to be an excellent proxy of district-level partisanship, particularly since the 1990s (Levendusky et al. 2008). There are a number of ways to look at legislative behavior for evidence of electoral responsiveness. In this section I draw on party unity scores. The Party Unity Support Score calculates the percent of times that a member votes with his or her party on party unity votes (defined when a majority of Democrats vote against a majority of Republicans). As mentioned in previous sections, CQ classifies votes as either party unity or bipartisan. Thus, the logic of this measure, applied to the argument that members are electorally responsive, is that members in competitive districts should have lower party unity support scores than members in safe districts. 7 The drawback of this score is that it does not capture the relative frequency of party unity votes in general. 5 I use the normal presidential vote rather than the member s own past vote share because of concerns with foresight on the part of the legislator as well as incumbency advantage. If a member foresees the importance of appearing as bipartisan or partisan, his vote share is likely to reflect that. In effect, assuming any foresight by members means that their vote share, even at time t-1, is endogenous to their behavior. An additional problem is the personal vote for incumbents. That is, a large margin of victory may imply a partisan district or it may imply that a member has created a strong personal brand, potentially because of their responsiveness to the district. 6 In all analyses in this paper, all districts (where data is available) are included. This includes districts that were redistricted. Ideally, instances where a district boundary was redrawn would be omitted, and the normal presidential vote in subsequent years would be omitted to drop the presidential year in the first year after redistricting. However, the author s current data on redistricting only includes an indicator for whether a district is in a state that was redistricted, meaning that removing these cases leaves minimal observations in the 93 rd (1972), 98 th (1982), 103 rd (1992), and 108 th (2002) Congresses. As a result, redistricted cases are included and are a source of measurement error. 7 A comparison has also been made using DW-NOMINATE scores, which include all non-unanimous (or highly lopsided) votes in the calculation of idea points. With these scores, the expectation is that members in competitive districts should have more moderate ideal points. The correlation between the normal presidential vote in a member s district and his/her NOMINATE score increases from 0.37 in the 93 rd House to 0.58 in the 103 rd House, but has declined in magnitude since the 104 th Congress. 9
Looking at the correlation between the normal presidential vote in the district and the party unity support score yields suggestive evidence of changes in representativeness over time (see Table 1). Although initially showing a relatively strong correlation with the normal vote, party unity support scores show a weakening relationship from the 97 th to 102 nd Congresses. [Table 1 about here] A second approach further parses out these patterns. I use a quasi-binomial model of legislative behavior, which accounts for the likelihood that a member engages in partisan (or bipartisan) behavior relative to the number of times they do not. Successes in the model are votes with the party. The primary independent variable is the Normal Presidential Vote in the district. I allow this effect, as well as the intercept, to vary by Congress. Finally, I include a number of individual level covariates member of the Majority Party, Female, Age, Tenure, and an indicator for whether the member holds a Leadership post (Speaker, Majority or Minority Leader, or Whip). For each of these control variables, the effects are constrained to be constant across time. A quasi-binomial, rather than a binomial model, is used in order to allow for overdispersion. The model is specified as follows: To assess the degree of responsiveness to the normal presidential vote over time I examine the predicted probability of partisan support over the range of the normal presidential vote for each Congress. The calculation of the predicted probability holds all variables in the model except the Normal Presidential Vote at either their mean or median. Figure 4 presents the predicted probability of a member voting with his party on party unity votes for selected 10
Congresses. 8 The 95% confidence intervals are included in the dashed lines. Over time, we see both an intercept shift toward greater party support as well as a decrease in the slope, with a low in the 101 st Congress. [Figure 4 about here] It is important to note that neither of these approaches provides a definitive benchmark of ideal representation. Rather, both of these approaches rely on comparisons over time within the measure, looking at either changes in the correlation or changes in the intercept and slope of the estimated relationship. In making the argument that these changes reflect declining representation, I am making the assumption that a higher correlation between the measures or a lower intercept and steeper slope reflect higher levels of representation. With these caveats in mind, both of these roll call patterns are suggestive that electoral responsive has declined. 3. Breaking Apart Legislative Behavior - Bill Cosponsorship Coalitions Although roll call votes are the standard measure of legislative behavior, and often for good reason, it is important to remember that roll call votes are not a random sample of legislation. Therefore, this paper considers a second measure of legislative behavior bill cosponsorship coalitions to understand patterns of partisan and bipartisan cooperation in Congress. Of course these are not the only two forms of bipartisan cooperation that can occur in Congress. Bipartisanship may also be observed in the processes used in Congress (i.e., the use of closed versus open rules or the composition of conference committees), or even in the rhetoric of members speeches. Nonetheless, roll call votes and bill cosponsorship coalitions are two areas that provide a unique opportunity to both systematically examine decisions by all members of Congress and to examine the influence of congressional agenda control. 8 The pattern over time is consistent in the Congresses not presented here and graphs of all Congresses are available upon request. 11
Despite the prevalence of roll call votes in Congressional literature, the bills that reach roll call votes are not a random sample of the bills or issues that have been brought up in Congress. Many scholars have argued that party leaders strategically put up issues for roll call votes that divide the parties from one another but do not show divisions within their own party (Snyder 1992; Poole 2004; Loewenberg 2008; Heller and Mershon 2008; Carrubba et al. 2008). If roll call votes are more likely on legislation that pit the parties against one another, this has the potential to bias analyses of roll call votes from finding high levels of bipartisanship. This is particularly true if legislation with bipartisan support passes via voice votes rather than roll call votes. Second, the likelihood that legislation, and particularly bipartisan legislation, faces a roll call vote may not be constant across time. For instance, Lynch and Madonna (2008), find that the prevalence of voice votes relative to roll call votes on significant legislation varies both across time and across issues. If the underlying data generation process for roll call votes is biased against bipartisan legislation, or if the data generation process varies over time in a way that is related to whether legislation has bipartisan support, then both the level of bipartisanship and the relationship between polarization and bipartisanship may be misleading. For instance, if the use of agenda control has changed over time, with bipartisan legislation being less likely to face a roll call vote, the analysis of roll call voting may be biased toward a reduction in bipartisanship over the same period. Because of these concerns, I explore bill cosponsorship coalitions as a second measure of bipartisanship in Congress. As noted by Krehbiel (1995) and Kessler and Krehbiel (1996), cosponsorship is less likely to be subject to agenda control. Like roll call votes, however, bill cosponsorship coalitions allow all members the opportunity to take a position and side with a 12
coalition of other members. By looking at bill cosponsorship coalitions, I am able to assess bipartisanship absent the strategic considerations of which bills face roll call votes. Section 3.1 Bipartisanship in Cosponsorship Coalitions Bill sponsorship has long been understood as an effective and relatively easy way for members of Congress to become involved in the policy process. Similarly, cosponsoring legislation carries a number of possible benefits, both within Congress and among constituents. As such, cosponsorship has become a frequent activity in both the House and Senate in recent decades. In 1967 the House passed a resolution allowing up to 25 cosponsors on a bill (Congressional Record 1967, 10708-12). Since it was first allowed in the House, cosponsorship has largely replaced the use of duplicate bills (Thomas and Grofman 1993), suggesting that members view cosponsorship in much the same way as they do sponsorship because they can make the same claims back in their constituencies. In the 95 th Congress, the House passed H. Resolution 86, which allowed unlimited numbers of cosponsors and allowed cosponsorship even after the bill was in committee (Thomas and Grofman 1993). Since the 1970s, members in both chambers have utilized cosponsorship frequently. There are a number of reasons to believe that cosponsorship coalitions are a useful stage in the legislative process to assess bipartisanship. First, cosponsorship provides House leaders with low-cost information about the political benefits of a bill (Koger 2003, 227). Second, what a member cosponsors, and with whom they cosponsor, allows members to send messages to their constituents. Even if a bill doesn't move, cosponsoring helps clarify your message. That way people know where you are... and that trickles down to constituents (quoted in Koger 2003, 232). 13
Though there have been debates about why members cosponsor, the assumption in this paper is that members cosponsor legislation that they genuinely favor (similar to a preference or matching argument). However, since members only cosponsor a fraction of the entire number of bills that they might favor, the selection of which bills to cosponsor (among those bills that they like) is assumed to reflect concerns for signaling (either to constituents or internally to Congress) or reciprocity among members. To measure bipartisan cosponsorship, I utilize cosponsor data collected by James Fowler (Fowler 2006a, 2006b) that provides a matrix of all bills and cosponsors in a given Congress. I use this raw data to create bill level measures of bipartisanship for all House bills. 9 Before focusing on bipartisan cosponsorship, it is important to understand the trends in more general cosponsorship patterns. On the whole, the use of cosponsorship has increased in the House since the 93 rd Congress (see Table 2). Whereas only one-quarter of House bills were cosponsored in the 93 rd Congress, since the 99 th Congress, more than half of all bills have been cosponsored. In the 108 th Congress, over seventy percent of bills were cosponsored. The median number of cosponsors on a House bill (including those that are not cosponsored) has also risen over time. Among bills that are cosponsored, the median number of cosponsors has nearly doubled in the period of analysis, from six in the 93 rd Congress to eleven in the 105 th through 108 th Congresses. However, members are selective in their cosponsorship. No more than six percent of all cosponsored bills have more than a hundred cosponsors. [Table 2 about here] Since cosponsorship has not been analyzed as extensively as roll call votes, there are no 9 I use only House bills throughout the analysis, omitting resolutions and amendments. Although all House bills are included in the analysis, omitting commemorative legislation, which coding is available for through 2002, does not change the results. The measures of bipartisanship with and without commemorative legislation are correlated at 0.98. 14
standard measures of partisan or bipartisan cosponsorship. Therefore, this paper considers a number of possible specifications. I begin by creating a dichotomous measure of bipartisanship where bills are bipartisan if they are above the lower bound of bipartisanship on the scale and are partisan otherwise. Taking the simplest definition of bipartisanship all cosponsored bills that are not unipartisan (where unipartisan refers to bills where all of the cosponsors are from the same party as the sponsor) suggests that bills have been and continue to be overwhelmingly bipartisan. When all cosponsored bills are considered, bipartisanship occurs on more than twothirds of cosponsored legislation. Looking at only bills with the median or more cosponsors, I find that an even higher proportion of these bills are bipartisan, and that although there is some decline over time, it is a decline from 0.94 to 0.82, a low that was reached in both the 94 th and 108 th Congresses under relatively low and relatively high levels of polarization, respectively. To prevent attributing bipartisanship to bills that merely have a single rogue member cosponsoring with members of the opposite party, I restrict my definition of bipartisanship to those bills that have at least twenty percent of the cosponsors from the party other than the original sponsor (see Figure 5). 10 When all House bills are considered, the proportion of bills that 10 Additional definitions of bipartisanship, including a raw number of cosponsors from the opposing party (5, 10, and 20) as well as different percentages of cosponsors from the opposing party (20%, 30%, and 40%) have also been used for comparison. In all cases the pattern over time is similar with the primary difference being a shift in the intercept. A potential concern about the analysis presented above is that those bills classified as bipartisan may not be that different from those bills that are classified as partisan, particularly since the inferences are often made from a small number of cosponsors. To verify that those bills classified as bipartisan are distinct from those bills classified as partisan in the analysis, I focus just on those bills that receive a roll call vote and examine the probability that each type of bill partisan and bipartisan by cosponsorship - receives an bipartisan roll call vote. That is, of all bipartisan (or partisan) cosponsored bills that face a roll call vote, what proportion end up having a bipartisan roll call vote (as defined by the CQ measure)? Using data from the Policy Agendas Project, Rohde s dataset of House roll call votes, and my bipartisan bill cosponsorship measures indicates that between one-quarter and one-half of bills with bipartisan cosponsors that reach roll call votes result in a bipartisan vote. The average for 1973 through 2000 (the last year in which all three data sources are available) is just about one-third. In contrast, between onetwentieth and one-fifth (with an average of three-twentieths) of bills with partisan cosponsors that reach roll call votes result in a bipartisan vote. Although it is not impossible for partisan cosponsored bills to result in a bipartisan roll call vote, it is rare. In all years, bills with bipartisan cosponsorship coalitions are more likely to result in a bipartisan roll call vote than bills with partisan cosponsorship coalitions. This suggests that the cosponsorship 15
are bipartisan increases over time, largely due to the overall increase in cosponsorship. Nonetheless, this is suggestive that bipartisanship has become an increasingly important tool for members despite the parallel rise in polarization. Between twenty and forty percent of all House bills are bipartisan. When the analysis is restricted to only cosponsored bills, I find that although there has been some movement and a slightly downward trend in bipartisanship over time, the magnitude of the change is surprisingly small. Over the entire period of the 93 rd to 108 th Congresses, the range of bipartisan cosponsorship is between 47% and 61%. The low point of 47% occurs in 1975, 1993, and 2004, under disparate polarization levels by standard roll call accounts. [Figure 5 about here] As done in the analysis of voting, I also create a difference in party support score for each bill. Since cosponsorship occurs only on the positive side, this is similar to looking at the Yea votes. For each cosponsored bill, I subtract the percent of Republican cosponsors from the percent of Democratic cosponsors. Like the roll call-based measure, the resulting the scale ranges from -100 to 100, where the endpoints reflect bills with all Republican cosponsors and all Democratic cosponsors, respectively. Looking at the density of the difference in party support scale, which ranges from -100 (purely partisan Republican bill) to +100 (purely partisan Democratic bill), for each House of Representatives from the 93 rd to the 108 th Congress indicates that bipartisanship persists. As seen in Figure 6, although the distribution of cosponsor coalitions is bimodal in the sense that there are a large number of both Democratic and Republican partisan bills, the middle (i.e., the most bipartisan bills) has not vanished over time. The primary change in the distribution of bills is that measures are capturing important variation, and that the importance of this variation extends to voting patterns of the chamber as a whole. 16
it has become less lopsided toward the Democratic side, particularly once the Republicans gained majority status. Whereas partisan bills used to be predominantly Democratic, there is a relative parity of Democratic and Republican partisan bills in more recent Congresses and a growth in moderately bipartisan bills on the Republican side. More on point for this analysis, however, is the finding that the center of distribution persists across time. If there is a decline in bipartisanship, we should have seen the density of both tails grow at the expense of the center, creating a greater U-shaped pattern over time. [Figure 6 about here] Using a continuous measure of the percent of cosponsors from the party opposite the party of the sponsor, reiterates the persistence of bipartisanship across time. This measure is similar to the party bipartisanship scale except that it directly accounts for the party of the bill s sponsor. Across the period of analysis, the mean percentage of bipartisan cosponsors fluctuates around 30%, indicating that, on average, one third of a bill s cosponsors are from the party opposite the party of the original bill sponsor. Regressing these various measures of bipartisanship on time reiterates the decline of bipartisanship on roll call votes but a greater persistence of bipartisanship on bill cosponsorship. Furthermore, since the correlation between time and polarization (using the difference in party means of Americans for Democratic Actions (ADA) scores) is 0.93, this approach sheds light on the extent to which (both voting and cosponsorship-based) bipartisanship has declined as votingbased polarization has risen. Beginning with the CQ measure of all roll call votes, the estimated beta on time is -0.45 (p < 0.05). The magnitude of the effect increases when only final passage votes are included, to -0.85 (p < 0.001). In contrast, for the 20% definition of bipartisan bill cosponsorship, the estimated beta on time is -0.25 (p < 0.01). Similar patterns are found when 17
bipartisanship is measured using the difference in party support scales. In this case, we expect the direction of the coefficient to be positive since lower absolute values of the scale reflect greater bipartisanship. Defining bipartisanship as the mean of the absolute value of the party scale, we get an estimated coefficient on time of 1.16 (p < 0.001) when looking at yea roll call votes (coefficient of 1.11, p < 0.001 for final passage votes), but a coefficient of -0.02 (p = 0.72) when looking at bill cosponsorship coalitions. Thus, regardless of the technique for estimating bipartisanship, bipartisan cooperation shows a greater decline over time in roll call votes than in bill cosponsorship coalitions. Section 3.2 Cosponsorship Coalitions and Electoral Responsiveness At the outset of this paper, I argued standard accounts of party polarization often imply a decline in electoral responsiveness. In the preceding section, I empirically demonstrated this declining, drawing on the roll call voting of members. In this section, I return to the assessment of responsiveness, following the same approaches but using bill cosponsorship coalitions rather than party unity support scores on votes. As in the previous section, district preferences are measured using the Normal Presidential Vote in the district. Focusing on bill cosponsorship patterns for the dependent variable, I define a member s propensity for Bipartisan Cosponsorship as the percentage of bills they cosponsored that are bipartisan. Bipartisan is defined using the definition where at least twenty percent of the cosponsors are from the party opposite the party of the sponsor. Looking at the correlation between the normal presidential vote in the district and bill cosponsorship behavior offers a stark difference from the patterns of roll call voting. As seen in Table 1, the normal presidential vote and members party unity votes showed a weakening relationship over time. In contrast, bipartisan cosponsorship measures show little to no 18
relationship for the first part of the series, but a growing correlation since the 102 nd Congress. Since the 103 rd Congress, the correlation between the normal vote and members bill cosponsorship coalitions has generally been larger in magnitude than the correlation between the normal vote and members party unity support scores. A second approach further parses out these patterns, again using a quasi-binomial model. In this model, successes are bipartisan cosponsorship coalitions. The primary independent variable and control variables are the same as in the analysis of roll call votes. Figure 7 presents the predicted probability of cosponsoring a bipartisan bill for a selected set of Congresses. As expected, the slope of the line is in the opposite direction than in the previous table since we have changed to looking at bipartisanship (rather than partisanship). More important, however, is the different pattern that we observe over time. Here, we see little movement in the intercept over time and an increase in the steepness of the slope. Both of these patterns suggest an increase in responsiveness to district preferences when looking at bill cosponsorship coalitions. [Figure 7 about here] Whereas members responsiveness to their districts declined in the roll call-based measure, it increased over time when using the bill cosponsorship measure. This result has significant implications both for the empirical study of Congress and representation as well as the normative interests in these questions. It is also worth noting that the increase in responsiveness of the cosponsorship measure corresponds time-wise to the low point of the roll call measure. Members of Congress may have realized that their roll call behavior placed them as out of step with their districts (potentially because of the types of bills that were selected to face roll call votes) and thus turned to other forms of legislative behavior to show responsiveness. In sum, this section demonstrates that members legislative behavior remains a function of their 19
district preferences but that where responsiveness occurs in the legislative process has changed over time. 4. Agenda Control and the Decline of Bipartisanship Thus far, the data have provided the following empirical patterns partisanship has increased in voting while bipartisan cooperation has declined, bipartisanship has changed relatively little on bill cosponsorship coalitions, and electoral responsiveness has declined when measured with roll call votes but increased when measured with bill cosponsorship coalitions. How can we reconcile these divergent patterns? One explanation is that changes in congressional agenda control over have led to the creation of an increasingly partisan roll call agenda. This change in agenda content may reflect decreases in overlap between the parties, combined with the benefits that come from creating distinctive party brands. This section presents an overview of this argument and some empirical evidence that supports this claim. Cox and McCubbins (2005, 5) suggest that the primary tool of the majority party is picking which bills will be voted on at all, and only secondary do they focus on garnering enough votes to get a victory. The party leadership is wary to put up issues for votes that create divisions within their own party. In contrast, the leadership often has incentives to place issues up for votes that are expected to divide the parties, with their party coming out on the winning side. A Congressional Quarterly staff writer notes that, The 2004 party unity scores reflect a continuing effort by party leaders to sharpen their distinctions and to rally their troops to vote in ways that highlight those distinctions (Poole 2004, 2906). He goes on to say that: Republican leaders followed a model they adopted in 2000, in which they carefully orchestrated what came to a vote and managed intraparty conflict to minimize dissent on the floor. In both the House and the Senate, there were fewer total roll call votes than in 2000, but the percentage of total votes that were party unity votes were somewhat higher in 2004 than in 2000. That is a reflection of the Republican leaders choreography (Poole 2004). 20
Political scientist Sarah Binder suggests that The rules of the game are easy enough to manipulate by a majority party to foreclose opportunities to vote on alternatives that would attract bipartisanship (quoted in Poole 2004). Electoral strategies geared at motivating the base to turnout can further heighten the disincentive for bipartisan agenda. Adam Nagourney, a reporter for the New York Times noted that when both sides are concerned with motivating their base, the agenda difference between the two is much more dramatic (Nagourney 2003). Likewise, Grynaviski (2010) argues that political parties perform the role of a surety to voters in elections, and are better able to perform this function when party unity is high. Although there is variation in whether members have electoral incentives to vote contrary to the party, he suggests that the use of agenda control limits the number of votes taken where there is an incentive to deviate from the party (Grynaviski 2010). Overtime, the sorting out of the political parties, resulting in reduced overlap between the ideologies of the members, has combined with agenda control to contribute to an increasingly partisan roll call agenda. This sorting is particularly apparent in the South, where Republicans replaced conservative Democrats. Sorting of the two parties, even without having the Democrats move left or the Republicans move right, creates opportunities for partisan legislation. As overlap declines, there is an increase in the number of status quo points for which an alternative exists that receives the support of the majority party median and the floor median, but does not receive the support from the minority party median. That is, there is an increase in the number of issues that meet Cox and McCubbins (1993, 2005) criteria for being placed on the agenda (support of the majority of the majority party) but that will result in party unity rather than a bipartisan vote. Thus, when the congressional parties were unsorted, as was the case during the heyday 21
of the Conservative Coalition between Republicans and Southern Democrats, votes often had a bipartisan makeup (Manley 1973). But since the 1970s, as conservative Southern Democrats were replaced by Republicans (Han and Brady 2007), overlap decreased, allowing partisan legislation to find room on the congressional agenda. As a result, legislation with bipartisan support became increasingly less likely to face a roll call vote. A second factor that may have made the roll call agenda increasingly partisan is the increased tracking of members voting records by interest groups, again beginning in the 1970s. When special interest groups are watching members behavior, the parties have greater incentives to select votes that pit the parties against each other, especially on those issues that the party relies on support or funding from an interest group. The fact that votes on those bills selected as most important by these interest groups tend to closely follow party lines attests to this dynamic (Snyder 1992). In order to explore how the incidence of bipartisanship changes when considering the cosponsorship stage versus the roll call vote stage of the legislative process, Figure 8 looks at the percentage of cosponsored bills that are bipartisan under the 20 percent definition for all House bills, for bills that reach roll call votes, and for bills that reach roll call votes and are classified as important. Two definitions of important legislation are used. The first uses Congressional Quarterly Almanac (CQ) and the second is original data that was collected matching up Mayhew s important legislation and George Edward s list of important failures with the appropriate bill numbers. 11 In each case, the graph plots the percent of bills in each category that are bipartisan in their cosponsorship coalitions. The lines are smoothed using a loess procedure 11 An updated list of Mayhew s important legislative enactments and George Edward s major legislative failures was graciously provided y Sean Theriault. The bill numbers of the specific piece of legislation, as well as duplicate bills and other similar proposals, was coded using CQ Almanac, the Thomas search engine from the Library of Congress, and relevant newspaper articles on the legislation if necessary. 22
that utilizes locally weighted polynomial regression. Here, higher percentages reflect greater bipartisanship. Comparing all cosponsored bills with bills that reach roll call votes, there is a greater decline in bipartisanship in those bills that face roll call votes than there is in overall bipartisan cosponsorship of legislation. Beginning in the mid-1990s, however, there is a resurgence in bipartisanship in bills that reach roll calls. The same pattern is seen when looking only at important bills (CQ) and even more apparent when considering major passages and failures. Combined, these findings suggest that the decline in bipartisanship evident in roll call voting is largely driven by the choice of which bills receive roll call votes. [Figure 8 about here] A second approach to distinguishing how agenda control may influence the measurement of bipartisanship is to look at the conditional probability of reaching a specific stage in the policy making process given that a bill is either partisan or bipartisan in its cosponsorship coalition. Using the 20% definition of bipartisan legislation, I analyze both the conditional probability of reaching a roll call vote as well as the conditional probability of becoming a public law given that a bill is bipartisan (or partisan) (Table 3). In both cases, bipartisan bills have a greater probability of reaching the legislative stage than partisan bills. However, I find different patterns over time in the conditional probability of reaching a roll call vote versus becoming a public law for bipartisan bills. Whereas the conditional probability of a partisan bill reaching a roll call vote slightly increases over time, the conditional probability of a bipartisan bill reaching a roll call vote generally declines from the 93 rd (early 1970s) to the 103 rd Congresses. The conditional probability that a bipartisan bill reaches a roll call vote increases only after the Republican takeover in the 104 th Congress. In contrast, the probability of becoming public law given that a bill is bipartisan has nearly tripled between the 93 rd and 108 th Congresses. This suggests that 23